add config and model.py

config.json

@@ -0,0 +1,26 @@
+{
+  "_name_or_path": "IDEA-CCNL/Ziya-LLaMA-7B-Reward",
+  "architectures": [
+    "LlamaRewardModel"
+  ],
+  "auto_map": {
+    "AutoModelForSequenceClassification": "rm_modeling_hf.LlamaRewardModel"
+  },
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_position_embeddings": 2048,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "pad_token_id": 0,
+  "rms_norm_eps": 1e-06,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float32",
+  "transformers_version": "4.28.0.dev0",
+  "use_cache": true,
+  "vocab_size": 32001
+}

modeling_llama_rm.py

@@ -0,0 +1,26 @@
+from transformers import PreTrainedModel,LlamaConfig,LlamaModel
+import torch.nn as nn
+import torch
+from typing import Optional
+
+class LlamaRewardModel(PreTrainedModel):
+    config_class =LlamaConfig
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = LlamaModel(config)
+        self.value_head = nn.Linear(config.hidden_size, 1)
+    
+    def forward(self,
+                input_ids: torch.LongTensor, 
+                attention_mask: Optional[torch.Tensor] = None) -> torch.Tensor:
+        outputs = self.model(input_ids,attention_mask=attention_mask, output_hidden_states=True)
+        last_hidden_states = outputs.hidden_states[-1]
+        if attention_mask is None:
+            last_hidden_states = last_hidden_states[:, -1]
+        else:
+            last_index = attention_mask.cumsum(dim=1).argmax(dim=1)
+            last_hidden_states = last_hidden_states.gather(1, last_index.view(-1, 1, 1).expand(-1, 1, last_hidden_states.size(-1))).squeeze(1)
+        values = self.value_head(last_hidden_states).squeeze(-1)# (bs,)
+
+        return values